The MOT should handle the fact that datasets are typically made of various datasets with different licenses

[lehors]

The MOT currently treats "Datasets" as a component with a single license when in practice the datasets used by model producers to train their models are made of a whole collection of datasets which come with various licenses.
Handling this in a user friendly way is however not obvious and deserves a bit of thinking.
This issue is here to record the current deficiency.